Text this: Distinguishing semantically similar queries in temporal video grounding via LLM-generated query.